verify-page-loaded

name	verify-page-loaded
description	Confirms that the current page has finished loading and is ready for interaction. Detects loading spinners, skeleton screens, and content placeholders. Run before extraction skills.
version	0.2.0
allowed-tools	["wait_for_dom_ready","wait_for_selector","wait","assert"]
metadata	{"applies_to":"any-website","url_patterns":["*"],"dom_markers":[],"flake_rate_target":0.02,"flake_rate_measured":null,"exercised_on":["wikipedia-list","arxiv-listing","bbc-home"],"cost_budget":{"deterministic_only":true,"max_vision_calls":0},"sensitive":false,"evaluate_success_criteria":true}

Verify Page Loaded

When to invoke

After any navigation, and after any interaction that may have triggered new content (clicking pagination, expanding a section, submitting a form). Especially important before extraction skills — extracting from a half-rendered DOM produces incomplete results.

This skill is vision-forbidden (max_vision_calls: 0). If the deterministic checks fail, the skill returns failed and the agent knows the page is genuinely not ready — vision wouldn't help.

Recipe

1. wait_for_dom_ready timeout=10s
2. wait_for_selector selector="body" state=visible timeout=2s
3. wait extra=200ms
4. assert no_visible_element selector="[aria-busy='true']" timeout=3s
5. assert no_visible_element selector=".loading, .spinner, [class*='skeleton']" timeout=3s

Success criteria

• assert dom_ready
• assert no_visible_element selector="[aria-busy='true']"
• assert main_content_present

The main_content_present assertion checks for one of: <main>, [role=main], <article>, or a body height > 200px (heuristic for "something rendered").

When NOT to use

• On pages that legitimately stay in a loading state (live dashboards, video-call interfaces, etc.). Use a custom skill or rely on agent reasoning.
• Inside an extraction loop where the page is the same as before; this skill assumes a fresh state.

Known failures

• Single-page apps with no DOMContentLoaded signal change: rare but happens with React/Vue apps that use shadow DOM heavily. Falls back to the spinner-absence check.
• Pages with always-present skeleton-style classes: some design systems use .skeleton for normal layout elements, not loading. Authors can override the recipe to remove step 5.
• Sites that block until consent is given: if a cookie banner is also blocking, run dismiss-cookie-banner first.

Example usage

Agent task: "Get the latest cs.AI papers from arxiv."
→ matcher selects [verify-page-loaded, extract-table-pagination]
→ invoke_skill("verify-page-loaded") → success in 89ms, deterministic_path=true
→ invoke_skill("extract-table-pagination") proceeds against a fully-rendered page

Related skills

• dismiss-cookie-banner — usually precedes this skill
• extract-table-pagination — usually follows this skill
• handle-infinite-scroll — uses similar primitives but with a different success criterion